!(https://hackernoon.com/hn-images/1*GBiknk1xvebj97qVfmks4g.png)\n\nLike many other critical business functions, machine [learning](https://hackernoon.com/tagged/learning) is slowly creeping into HR departments. The buzz of machine learning has some people thinking that it could eliminate all of our jobs, while it has others thinking it could be the next big wave of productivity. The truth is, one won’t replace the other — [technology](https://hackernoon.com/tagged/technology) and people will ultimately need to co-exist because machines can’t do it all.\n\nFor example, in a recent [Talk Talent to Me](https://www.talktalenttome.com/lightspeed-ventures-talent-partners-podcast/) podcast episode, Cat Surane and Luke Beseda at Talent Partners for Lightspeed Ventures discussed the importance of understanding a candidate’s motivations for pursuing a new job — are they just looking for a higher salary? Are they using your job offer as a bargaining chip with their current employer? Machines aren’t good at evaluating the emotions driving a job search; you need a human to do that.\n\nWith that said, machine learning can do a lot to make recruiting more efficient, which is desperately needed given that it takes 43 hours on average to recruit one engineer. Here’s what machine learning should do:\n\n* **Provide recommendations**. Today, the majority of HR professionals leveraging recruitment platforms are sourcing talent via a search based system where profiles are surfaced based on factors such as location, skills, experience and industry. Since this doesn’t take into consideration whether or not the candidate will be a good culture fit, it actually ends up making the process more challenging. That’s where machine learning comes in, shifting hiring tools from a sourcing method to a recommendations paradigm. Instead of sifting through a list of 500 candidates, job and networking sites can use machine learning to intelligently suggest candidates who have the right qualifications for the specific open role. Technological developments that provide deeper visibility into the candidate pool help job seekers and recruiters use their time more efficiently to find the best job fit, which is exactly what machine learning should be doing for recruiting.\n* **Understand unstated goals**. Classic job descriptions have long been criticized for being too generic or all-encompassing. Machine learning can help by uncovering candidates that match an organization’s unstated goals, such as improving workforce diversity. When unstated factors are known, HR teams reap the benefits due to high-powered search capabilities that identify tailored candidate matches. \n Machine learning can also play a major role in addressing tech’s gender wage gap. For example, it can source and analyze salary data to help companies understand what they should be offering candidates, based on their specific skills and experience, not their previous salary.\n\n**Machine learning shouldn’t:**\n\n* **Bias against diversity.** Organizations that are considering using machine learning to improve their hiring process will likely find that building an equitable platform, free of unbiased hiring and wage gaps, will be the biggest challenge that they face. When employing machine learning, organizations need to ensure that the dataset being used to train the algorithm is normalized to remove all biases. \n Normalizing data means adjusting values measured on different scales to a common scale. Lets use an example to understand this. If we have 1000 candidates in our dataset — 800 men and 200 women — and this data is used to determine who will have a successful outcome, the algorithm is going to most likely predict men having successful outcomes, just because the volume of men in this dataset is far greater than women. Normalizing this data means taking 200 women and 200 men and using this normalized dataset to build the algorithm. This is critical for the tech hiring industry in particular because if you have more volumes of inherent data in an algorithm then the algorithm will be biased; i.e., if we were to not normalize that data, over time women wouldn’t even make it onto the platform.\n\n!(https://hackernoon.com/hn-images/1*S13HwBZJovEWINA2U3ELNg.png)\n\nThere’s more that goes into whether or not someone accepts a job or is suited for a particular role than what can be analyzed in a system. At a recent [SourceCon, Glen Cathey](http://booleanblackbelt.com/2012/01/talent-sourcing-man-vs-aiblack-box-semantic-search/) talked about where technology can’t replace humans. Cathey claims that anything that falls under the umbrella of context, nuance, or empathy should not be threatened by AI.\n\nTrust is another important factor. Accepting a job offer is a hugely personal and important decision, and having a machine manage the entire process doesn’t fill a person with trust. Human interaction ensures candidates feel that their priorities, whether it a salary request or a concern about team structure, are being clearly communicated and considered.\n\nOrganizations that will be successful with leveraging machine learning in hiring are the ones that use it to advance productivity, not do their job for them. To fix the broken process recruiters will need to add value during the interview prep, offer assistance with negotiations and continue to be an advocate for both parties involved. At the end of the day, humans and algorithms need to work together, and if leveraged correctly, machine learning will give an immediate boost to the industry.